diff --git a/docs/paper-c.adoc b/docs/paper-c.adoc index 1c1160c..f4b665b 100644 --- a/docs/paper-c.adoc +++ b/docs/paper-c.adoc @@ -22,9 +22,15 @@ discipline across three tiers — node, server, and browser — and find it effectively free: whole-file hashing adds ~0.1 s against 7-13 s session creation, verified reloads complete in ~0.4 s from cache, and a 202 MB int8 model runs end-to-end in a headless browser over WASM. -A margins-based quantization gate (keeping the 1.2M-parameter head in -fp32) cuts silent int8 flips 36x at +0.4% artifact size. Eleven-to-17 -models ship under the contract with full provenance. +Since the benchmark round, a depth-cut student (encoder 12->6, +verbatim layer copy) ships a ~95 MB int4 tier — full utility at +browser-native weight (0.0364% argmax flips, all near-tie; KLD +1.2e-4). A margins-based quantization gate (keeping the +1.2M-parameter head in fp32) cuts silent int8 flips 36x at +0.4% +artifact size. The live surfaces: the demo at +https://interscript.org/neural (lead artifact: the 95 MB lite tier), +the REST surface at https://api.interscript.org, and the +tag-pinned index channel (index-v3). == 1. Introduction - the deterministic-map discipline (OGC ISC context) and why learned